Journal of Monetary Economics 90 (2017) 8498

Contents lists available at ScienceDirect

Journal of Monetary Economics

journal homepage: www.elsevier.com/locate/jmoneco

Asset price volatility, price markups, and macroeconomic

uctuationsR
Miguel A. Iraola, Manuel S. Santos
School of Business Administration, Department of Economics, University of Miami, Coral Gables, FL 33146 United States

a r t i c l e i n f o a b s t r a c t

Article history: A variant of the neoclassical growth model is considered to study the role of innovation,
Received 14 November 2014 lags in technology adoption, total factor productivity TFP, and price markups as main de-
Revised 4 July 2017
terminants of asset price volatility. The model confers a prominent role to price markups
Accepted 5 July 2017
as opposed to other macroeconomic sources of uncertainty. In the data, price markups are
Available online 12 July 2017
highly correlated with stock market values, whereas other nancial measures of protabil-
Keywords: ity exhibit much less volatility and are weakly correlated with stock market values.
Technological innovations 2017 Elsevier B.V. All rights reserved.
Price markups
Stock market volatility
Price-dividend ratio
Taxes

1. Introduction

A burgeoning literature in the frontier between economics and nance has emerged to study asset price volatility, but
there is no general consensus on the driving forces of stock markets which remain a puzzle to economists. A variant of
the neoclassical growth model is here proposed to analyze technology innovation, TFP shocks, and price markups as main
determinants of stock market volatility. Conceptually, the aggregate asset value in our model should be identied as the
market value of corporations (MVC): the sum of the market values of corporate equity and net debt. The joint consideration
of these two components avoids the introduction of arbitrary policies for corporate debt and dividends.1 Payouts to debt
holders have been fairly erratic in recent decades (e.g., Hall, 2001). The volatility of MVC is mainly driven by stock values:
the volatility of MVC is about 6% below that of equity. (In our data the volatility of equity is about 26.40% and the volatility
of MVC is about 24.93%.) The volatility of MVC will be related to various macroeconomic aggregates as well as to the stock
market return and the price-dividend ratio.
Our analysis focuses on medium and long-term volatility. That is, uctuations of economic aggregates over a frequency
band of 250 years. This volatility should be easier to study as it is isolated from noisy information. Still, the economics
literature has struggled to come to terms with pronounced uctuations in stock market values commonly known as the
excess volatility puzzle.
First, the observed volatility of stock market values and price-dividend ratios is about ten times greater than that of
output and consumption, and about three times greater than that of real investment (e.g., see Gomme et al., 2011). Dynamic

R
We are very grateful to a referee of this journal for very constructive comments and suggestions.

Corresponding author.
E-mail address: msantos@bus.miami.edu (M.S. Santos).
1
Akin measures for the market value of US corporations were considered by Hall (2001); McGrattan and Prescott (2005), Peralta-Alva (2007), and Wright
(2004).

http://dx.doi.org/10.1016/j.jmoneco.2017.07.002
0304-3932/ 2017 Elsevier B.V. All rights reserved.
 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498 85

equilibrium models have failed to account for the volatility of stock values based on macroeconomic uncertainty even
after allowing for agent heterogeneity and market frictions. Second, asset market volatility is also disconnected from the
real economy (Alburquerque et al., 2016; Greenwald et al., 2014), and nancial measures of protability are weakly correlated
with stock values. This poses a challenge for traditional consumption-based asset-pricing theories. And third, stock market
returns are much more volatile than short-term interest rates. Most of the volatility of stock market returns stems from
unexpected changes in future excess returns (Campbell and Ammer, 1993). Several models can generate reasonable equity
premia, but are unable to account for plausible levels of asset price volatility. Our research should then complement theories
of the equity premium.
Our model is a simplied variant of those in Romer (1990) and Comin and Gertler (2006). Technological innovations
arrive exogenously to the economy and undergo a process of adoption embedded in the production of new varieties of
intermediate goods. Changes in price markups arise exogenously, which may be due to uctuations in economic conditions
and variations in the elasticity of substitution or monopoly power after the creation of new goods. Episodes of technology
innovation and shocks to productivity and price markups may produce persistent uctuations in the aggregate value of
stocks and R&D expenditure. Similar propagation mechanisms are considered in Grleanu et al. (2012), who include large
infrequent technological innovations embodied into new capital vintages. In contrast to these authors, we carry out a general
equilibrium analysis of the volatility of nancial variables along with other macroeconomic aggregates.
In our numerical experiments most of the volatility of nancial variables comes from price markups, whereas technology
innovation, lags in the adoption of new technologies, and TFP shocks may only have a signicative impact at the expense
of implausible uctuations in the real economy. Our model generates about two-thirds of the observed volatility in the
nancial variables. Movements in the stock market value and the price-dividend ratio remain fairly isolated from changes in
output, consumption, and investment.
Several recent contributions consider long-run productivity risks (e.g., Croce, 2014; Kaltenbrunner and Lochstoer, 2010;
Kung and Schmid, 2015) and time-varying nancial risks (e.g., Alburquerque et al., 2016; Gourio, 2012). While all these papers
get sizable equity premia and term premia they consistently fail to capture the long-term volatility of stock prices. Markups
and technology innovation are of interest to account for the cross-section distribution of returns and the value premium (e.g.,
Gomes et al., 2003; Kogan and Papanikolaou, 2014).
The paper will proceed as follows. Section 2 documents that markups are positively correlated with stock market prices,
and display great variation across company cohorts and over time. Section 3 lays down our two-sector model with technol-
ogy adoption. Section 4 tests the performance of the model on the volatility of nancial variables along with other macroe-
conomic aggregates. The various sources of volatility of the price-dividend ratio are presented in Section 5. We conclude in
Section 6.

2. Stock market volatility and markups

As already pointed out, to assess nancial asset volatility our analytical framework relies on the market value of cor-
porations (MVC). This is a consolidated measure of the market values of equity and net debt liabilities. Two data sources
are considered. First, in this section all data are taken from US companies in Compustat North America over the 19502012
period. This will allow us to compare the volatility and correlation of MVC with various protability measures at the mi-
cro level (i.e., across companies). Second, following Hall (2001) and McGrattan and Prescott (2005) we use the Financial
Accounts of the US over the 19482007 period as our primary source to map our model to the data. This latter database
incorporates both publicly and privately held corporations, which may pick up potential impacts of recently founded compa-
nies. Thus, companies are included since their foundation dates, e.g., Google was founded in 1998 and went public in 2004.
Hence, Google is included in our database since 1998. In both data sets, the volatility of MVC is roughly ten times greater
than that of output and consumption, and MVC is hardly correlated with output and consumption. Moreover, MVC is also
uncorrelated with our measures of dividends, stock market returns, and interest rates, but it is fairly correlated with the
price-dividend ratio.
Fig. 1 decomposes MVC for different company cohorts in the recent IT revolution. Market capitalization relative to aggre-
gate corporate value added is broken down into the following groups of companies: (i) Firms listed before 1970, (ii) Firms
listed in 19701979, (iii) Firms listed in 19801989, and (iv ) Firms listed since 1990. As one can see, most added value
belongs to new corporations. Next, the nancial performance of these cohorts will be assessed over various measures of
protability.
The nance literature has introduced several cash-ow and protability measures to study stock-return predictability.
The most common ones are dividends D1 and net income NI. As shown in Barsky and DeLong (1993), both measures are
reasonably correlated with stock market values at low frequencies. Although Fama and French (2006) nd that NI has some
predictive power for cross-section company returns, Novy-Marx (2013) argues that gross protability GP seems more appeal-
ing. GP and earnings before extraordinary items IB are usually called clean accounting measures. Some popular protability
measures like EBITDA, EBIT, and operating income before and after depreciation, OIBDP and OIADP, may also be of some
interest. All these nancial measures are detailed in the Appendix.
Table 1 offers a broad evaluation of rms nancial performance. For each company cohort, this table is intended to
compare the relative weight of MVC against commonly used nancial measures by considering averages over three different
86 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498

4.5

4
All Firms

3.5
Firms Listed Before1990

3 Firms Listed Before 1980

2.5
Firms Listed Before 1970

1.5

0.5

0
1950 1960 1970 1980 1990 2000 2010
Fig. 1. Market value of corporations over corporate value added for various groups of rms.
Notes: See Appendix for denitions and data sources.

Table 1
Markups and shares of MVC and nancial accounting measures across company cohorts.

Year: 9094 9599 0004

Cohort: 70 7080 8090 70 7080 8090 9000 70 7080 8090 9000

MVC 66.70 11.24 19.98 57.97 12.00 18.35 11.68 51.00 13.77 17.81 16.62
MU50 80.44 73.42 158.92 57.69 56.42 87.07 164.71 43.26 41.93 70.07 121.47
MU75 98.87 88.31 121.68 94.30 83.67 104.18 168.16 88.70 74.98 101.35 165.62
MU100 97.83 90.26 123.32 98.38 89.18 109.55 116.33 97.95 95.35 86.57 108.71
MU 94.51 106.60 121.25 97.43 110.84 113.50 91.45 98.98 114.66 122.37 83.84
D1 69.88 12.29 16.30 65.50 11.47 13.12 9.90 60.20 9.86 12.85 14.78
NI 67.06 17.25 14.37 65.10 13.13 14.86 6.90 190.17 47.63 24.18 159.86
GP 67.74 11.73 18.49 57.59 11.53 17.24 13.63 50.92 10.65 17.51 19.60
IB 69.36 14.94 14.43 64.59 13.82 15.42 6.17 102.38 25.42 25.29 51.76
EBITDA 65.30 13.66 19.28 56.81 14.05 18.00 11.13 50.62 12.93 19.13 15.97
EBIT 62.70 15.15 20.09 55.23 14.92 18.18 11.66 51.62 13.75 19.41 13.98
OIBDP 65.30 13.66 19.28 56.81 14.05 18.00 11.13 50.62 12.93 19.13 15.97
OIADP 64.33 14.52 19.17 56.44 14.34 17.86 11.36 52.13 13.41 19.45 13.78

Notes: Reported values are averages for the various time intervals. MVC: market value of corporations. MU50, MU75, MU100: average markup for the top
50%, 75%, and 100% companies with the highest ratio of R&D expenditure over total revenue, respectively. MU: average markup. D1: dividends. NI: net
income. GP: gross prot. IB: earnings before extraordinary items. EBITDA: earnings before interest, taxes, and depreciation. EBIT: earnings before interest
and taxes. OIBDP: operating income before depreciation. OIADP: operating income after depreciation. See Appendix for denitions and data sources.

time intervals: 19901994, 19951999, and 20 0 020 04. The price markup of a company is dened as the ratio between
total revenue and total variable cost, i.e., entries REVT and COGS in the Compustat data set. The aggregate price markup is
then obtained as a weighted average of company markups where the weights are the shares of company revenues. To obtain
markup estimates, companies are ranked by R&D intensity at every given date. MU50 refers to the average markup of the
top 50% companies with the highest ratio of R&D expenditure over total revenue, MU75 refers to the average markup of
 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498 87

Table 2
Correlation coecients and relative volatilities.

Time interval: 1 3 5 7 10 20 30 VOL

MU50 0.00 0.35 0.50 0.65 0.70 0.83 0.74 0.62

(0.09) (0.11) (0.10) (0.09) (0.09) (0.06) (0.08)
MU75 0.14 0.36 0.39 0.49 0.51 0.72 0.80 0.21
(0.11) (0.13) (0.14) (0.13) (0.15) (0.11) (0.06)
MU100 0.46 0.46 0.53 0.62 0.68 0.82 0.80 0.21
(0.14) (0.14) (0.13) (0.12) (0.14) (0.08) (0.06)
MU 0.14 0.18 0.08 0.07 0.06 0.07 0.08 0.43
(0.09) (0.10) (0.14) (0.17) (0.20) (0.10) (0.17)
D1 0.14 0.05 0.25 0.50 0.51 0.56 0.42 0.31
(0.14) (0.12) (0.13) (0.14) (0.13) (0.13) (0.16)
D2 0.12 0.01 0.19 0.15 0.15 0.11 0.57 0.37
(0.12) (0.13) (0.13) (0.16) (0.20) (0.13) (0.13)
NI 0.35 0.29 0.13 0.05 0.07 0.08 0.15 1.38
(0.12) (0.11) (0.14) (0.16) (0.18) (0.17) (0.15)
GP 0.47 0.35 0.21 0.02 0.11 0.42 0.55 0.34
(0.15) (0.21) (0.19) (0.20) (0.19) (0.12) (0.10)
EBITDA 0.46 0.38 0.23 0.01 0.08 0.33 0.38 0.39
(0.13) (0.18) (0.21) (0.24) (0.22) (0.14) (0.10)
EBIT 0.36 0.36 0.27 0.12 0.10 0.17 0.01 0.58
(0.13) (0.15) (0.17) (0.20) (0.21) (0.17) (0.12)
OIBDP 0.46 0.38 0.23 0.01 0.08 0.33 0.38 0.39
(0.13) (0.18) (0.21) (0.24) (0.22) (0.14) (0.10)
OIADP 0.40 0.36 0.19 0.05 0.11 0.23 0.25 0.52
(0.14) (0.17) (0.20) (0.22) (0.22) (0.17) (0.09)

Notes: Contemporaneous correlations for log differences of MVC with markups and nancial accounting measures over different time intervals. These
growth rates for MVC and the nancial accounting measures (excluding markups) are scaled down by growth of aggregate corporate value added. The last
column, VOL, is the ratio of the standard deviation of each variable over the standard deviation of MVC for a frequency band of 250 years. MVC: market
value of corporations. MU50, MU75, MU100: average markup for the top 50%, 75%, and 100% companies with the highest ratio of R&D expenditure over total
revenue, respectively. MU: average markup. D1: dividends. D2: corporate value added less investment, wages, and taxes. NI: net income. GP: gross prot.
EBITDA: earnings before interest, taxes, and depreciation. EBIT: earnings before interest and taxes. OIBDP: operating income before depreciation. OIADP:
operating income after depreciation. Newey-West corrected standard errors are in parentheses. See Appendix for denitions and data sources.

the top 75% companies with the highest ratio of R&D expenditure over total revenue, MU100 refers to the average markup
of all the companies reporting R&D activity, and MU refers to the average company markup in our sample.2 Markups are
reported as percentages of the average markup in the corresponding date. Hence, a value above 100 means that this cohort
can secure a higher markup than the sample average. For every other measure, the table reports the percentage or relative
value belonging to this cohort. For instance, on the rst column under the dividends D1 entry of the table one can read
that the value is 69.88. This means that the companies originating before 1970 were able to secure 69.88% of the total sum
of dividends in the sample of companies over the 19901994 period, whereas they only represent 66.70% of total MVC. The
youngest cohorts command the highest markups, but appear to be indistinguishable in terms of every other protability
measure.
This lack of correlation at the cross-section level is also validated in a further time series analysis for our sample of
companies over the 19602012 period.3 Table 2 reports correlation coecients between log differences of MVC and log dif-
ferences of every aforementioned variable over various time frequencies. This table incorporates a new measure of aggregate
income for shareholders and bondholders, D2, which is computed by taking investment, wages, and taxes out of aggregate
corporate value added and so it is inclusive of interest payments to debt holders. Similar measures have been considered by
various authors (cf. Boldrin and Peralta-Alva, 2009; Larrain and Yogo, 20 08; McGrattan and Prescott, 20 05). Growth rates for
MVC and the nancial accounting measures (excluding markups) are adjusted by the growth of aggregate corporate value
added. While markups may display correlation coecients of about 0.80 over the range of ve- to thirty-year frequencies,
nancial accounting measures such as D1 display correlation coecients of about 0.10. Most correlation coecients of the
protability measures are not statistically signicant at conventional condence levels, and their volatilities are rather small;
see last column of Table 2.
Similar calculations with sectoral data are performed to check for the robustness of these ndings. Sectors are
dened following Fama and Frenchs ve-industry classication codes (Cnsmr, Manuf, HiTec, Hlth, and Other) from
Kenneth Frenchs web page. Two further categories for high-tech companies are identied: (i) Companies whose stocks

2
Companies performing R&D activities in our sample represent about 39% of total market capitalization. The group of companies in MU50 represents
about 18% of total market capitalization, and the group of companies in MU75 represents about 25% of total market capitalization. Several other methods
have been proposed to measure markups (cf. Nekarda and Ramey, 2013).
3
Because of our limited sample of companies in the Compustat data set at early dates, the 19501960 period is taken out of the sample.
88 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498

are currently traded in the NASDAQ stock exchange, and (ii) Companies with SIC codes 281, 283, 284, 289, 357, 367, 381,
and 384. Nearly all correlation coecients between the log differences of MVC and the log differences of MU50, MU75,
and MU100 over ve- and ten-year time intervals are statistically signicant. Correlations over the ten-year frequency are
slightly stronger with coecients ranging between 0.60 and 0.80. Therefore, the robust correlation between changes in MVC
and markup measures extends to sectoral data.
The average price markup of the economy MU does not mimic so well the evolution of MVC. Our markup measure
MU50 usually displays the highest correlations with stock market values. It may be possible to construct some other related
protability measures over selected company groups that are highly correlated with stock market values.4 But price markups
can be microfounded and have become a basic ingredient of New Keynesian models. Markups may arise from optimal
pricing rules and price rigidities, and can be related to changes in elasticities because of income and substitution effects,
economic policies, and technology adoption.

3. A simple model of technology adoption

The economy is populated by a continuum of identical households. The aggregate consumption good is produced by a
single rm under a constant-returns-to-scale technology. Three inputs are involved in nal production: capital accumulated
by the rm, labor, and a composite intermediate good of technological products. Both rm and consumers act competi-
tively in all markets, but the sector of intermediate goods is composed of monopolistic producers. The range of available
intermediate goods can be expanded by a xed set of local adopters upon the arrival of new technologies.

3.1. The household

At each date t = 0, 1, . . . , the representative household supplies one unit of labor inelastically. Preferences are represented
by the following Epstein-Zin recursive formulation:
   11  11
1
Ut = (1 )ct1 + Et Ut+1 (1)

where Et is the expectations operator at t, 0 < < 1 is the subjective discount factor, 0 < = 1 is the inverse of the elasticity
of intertemporal substitution, and > 0 is the coecient of relative risk aversion.
The agent may participate in nancial markets by trading shares at of an aggregate stock. Let qt be the price of a unit
share, and dt denote the stochastic dividend. For given stock prices qt , exogenous wages t , lump-sum transfers Tt , and
initial asset holdings a0 , the optimization problem is to choose a stochastic sequence {ct , at+1 }t0 to attain the maximum
utility in (1) subject to the sequence of budget constraints:
ct + qt at+1 = t + (qt + dt )at + Tt (2)
where ct 0 and qt at+1 0, for all t 0.

3.2. The production sector

The rm producing the nal good accumulates capital and buys labor and intermediate goods. Final production Yt is
determined by a CES function subject to a TFP shock represented by a random variable At . At every date t there is a mass
t of intermediate goods that enter into the production of the nal good. These intermediate goods are bundled together

1
in a composite good Mt dened by a CES function Mt = [ 0 t ms,tt ds]t where ms,t denotes the amount of intermediate good
s bought by the rm at time t, and t > 1. Giveninitial condition k0 , the rm chooses stochastic sequences of investment,
labor, and intermediate goods {it , lt , (ms,t )s[0,t ] so as to maximize the present value of dividends
t0



E0 t dtf (3)
t=0

subject to

 t 
dtf (1 s )Yt it + t lt + ps,t ms,t ds c AEt TtLS (4)
0
   1
Yt At kt lt1 + (1 )Mt , 0 < , < 1, < 1 (5)

4
On related research, Gabaix (2011) argues that investment decisions of the top capital spenders have a big impact on aggregate uctuations. Capital
expenditures of the top 100 rms make up for over 60% of the aggregate investment of the publicly traded rms (Grullon et al., 2013). Using information
on US public rm patenting, Bena and Garlappi (2013) document that in a given year about 303 rms are actively innovating, and these innovating rms
account for about 40% of total market capitalization.
 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498 89

 t

AEt (1 s )Yt t lt + ps,t ms,t ds + a kt + T ded (6)
0

kt+1 = (1 )kt + g(it /kt )kt . (7)

Here, t is a state price converting income of period t to period 0, and ps, t is the price of intermediate good s at time t.
Dividends in (4) include a sales tax s , a corporate tax c over accounting earnings AEt , and a lump-sum tax TtLS lsY Nt
dened as a constant fraction of aggregate value added YNt . Taxable accounting earnings allow for capital depreciation a kt
and additional tax deductions dened as a constant fraction of the steady-state aggregate value added: Tded ded YN. Taxes
are rebated back to the representative household as a lump-sum payment, Tt . The capital stock kt depreciates at a constant
rate 0 < 1. Physical investment i is subject to adjustment costs (Jermann, 1998):
1
1 1
i
g(i/k ) = + (8)
1 1 k 1
where > 0 is the elasticity of the ratio i/k over Tobins q.
Forcing variables At and t are governed by stochastic processes:
ln(At ) = A ln(At1 ) + A tA (9)

ln(t ) = 0 + 1 ln(t1 ) + t1
(10)
iid iid
where A , 1 (0, 1 ), A > 0, tA N (0, 1 ), and ln(t ) N ( , ).
Monopolistic competition prevails in the technology sector for intermediate goods. Each variety s is supplied by a single
producer under a linear cost function c (m ) = m. Producer of variety s picks an optimal pricing strategy ps,t based on quantity
ms,t after assuming a xed set of prices and quantities for all other varieties. More precisely, for each time period t producer
of variety s maximizes the amount of after-tax prots:
s,t max {(1 c )( ps,t ms,t ms,t )} (11)
ms,t 0

where ps,t should be viewed as a function of ms,t from the inverse demand
 m  1t
s,t t
ps,t = pt (12)
Mt
  11
of the aggregate rm, and pt = ( 0 t ps,t t ds )1t .
Production of intermediate goods may be discontinued because of exogenous factors. Let be the probability of survival
of a technology at every date t. Let Vs,t be the present value of operating technology s from the beginning of time t:
 
 r rt
Vs,t = Et s,r . (13)
r=t
t

3.3. Technology adoption

The average stock of technological innovations Zt evolves according to the law of motion
Zt = Zt1 + xt1 (14)
with normalizing constant > 0 and
l nxt = x l nxt1 + x tx (15)
iid
where x (0, 1 ), x > 0, and N ( 0, 1 ). tx
Technologies are put into use by local adopters. The adoption sector is composed of a continuum of agents i [0, 1] that
behave competitively. Each adopted technology sells at price Vt to a monopolistic producer of intermediate goods. Let ti be
the stock of already adopted technologies by agent i, and (Hti ) the probability of adopting a new technology after investing
the amount of resources Hti . The stock t+1
i follows the law of motion:
 
t+1
i
= (Hti ) Zti ti + ti . (16)
The optimal Hti is derived from the following Bellman equation in which the value function is the option value Jti of a new
technology:
   i 
Jti = max Hti + Et (Hti )Vt+1 + 1 (Hti ) Jt+1
t+1
. (17)
Hti t
Observe that the optimal amount of expenditure Hti is the same for all i. Without loss of generality, we assume Zti = Zt for
 i
all i, and consider the aggregate stock of adopted technologies t+1 = t+1 di.
90 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498

3.4. Equilibrium and asset prices

For the nal good, market clearing holds if

Y Nt Yt t mt = ct + it + Ht (Zt t ) (18)
where YNt denotes aggregate value added, t mt is the cost of producing the composite intermediate good, and mt is the
amount produced of each variety.
The next proposition is central to our study. Let us dene the aggregate dividend dt dt + t t Ht (Zt t ). Assume
f

that the aggregate supply of the asset at = 1 for all t 0. Then, from the rst-order conditions of the representative house-
hold, one can obtain the following decomposition of stock market values:

Proposition 3.1. The stock market value:

qt = pkt kt+1 PVTt + Vt+ t + Jt+ (Zt t ) + t (19)

where
 
1

 r LS  
r
pkt , PV Tt Et T T
c ded
, Vt Vt t , Jt Jt + Ht , and t Et
+ +
J (Z Zr1 ) .
g
( kitt ) r=t+1
t t r=t+1
t r r

Therefore, the stock market incorporates the value of installed capital pkt kt+1 , the present value of lump-sum taxes PVTt ,
the value of adopted technologies Vt+ t , the option value of inventions currently available but not yet adopted Jt+ (Zt t ),
and the present value of future inventions expected to happen t . These latter components are further sources of asset price
volatility.
From numerical experimentation, in all calibrations below the model appears to have a unique ergodic invariant distri-
bution. It seems then adequate to simulate the model using a high-order perturbation method (Schmitt-Groh and Uribe,
2004) that takes into account the high volatility of stock market values. To check accuracy for the computed solution, this
approximation method is combined with a numerical dynamic programming algorithm (Santos, 1999) for the computation
of Bellmans Eq. (17).

4. Numerical experiments

This section presents an easy-to-follow calibration with uncorrelated shocks as well as an extensive sensitivity analysis
towards understanding the role of lags in technology adoption and stochastic markups. A good part of the variation of nan-
cial variables can be ascribed to uctuations in exogenous markups, and hence the calibration of the law of motion (10) is
most critical. Our tax structure is intended to match some rst- and second-order moments observed in the data. Without
taxes, the share of dividends in output will be too high. As before, denitions and data sources are gathered together in
the Appendix. As in D2 above, dividends are dened from corporate value added after taking out investment, wages, and
corporate taxes. All values refer to annual data from 1948 to 2007.

4.1. Baseline calibration

The second column of Table 3 lists parameter values for our baseline calibration BL. Parameters and are set to 5,
corresponding to the CRRA utility function, which falls within the range of empirical estimates for many studies. Parameter
= 0.95, leading to an annual interest rate of 5.26%.
Parameter is set to 0.2, which implies a labor income share in aggregate value added equal to 0.63 in the deterministic
steady state. From data in the manufacturing sector, the share of materials in nal production is generally assumed to be
around 0.50 (e.g., Comin and Gertler, 2006; Jaimovich and Floetotto, 2008). Letting = 0.65, this share becomes equal to
0.54 in the deterministic steady state. Our model matches the average investment to capital ratio in the data under an
annual depreciation rate = 0.09.
For the process of technology adoption, our calibration reproduces the volatility and persistence of the expenditure pro-
cess H (Z ) rather than the increment in the stock of adopted technologies (i.e., t+1 t ) which is harder to measure.
Barlevy (2007) shows a high degree of correlation between R&D spending from the NSF and Compustat data. We use the
NSF data. Following Hall (2007), the survival rate of each intermediate product is set to 0.98. The probability of adoption
is determined by a simple function

(Ht ) = 0 + 1 Ht (20)
with 0 , 1 > 0 and (0, 1). Parameter 0 > 0 is xed so that the steady-state value for probability (H) is equal to 0.166,
i.e., an average adoption time of six years. Thus, our calibration imposes much less persistence than other studies such
as Comin and Gertler (2006) in which this mean value equals 0.10. Parameters and 1 along with the law of motion in
(15) are adjusted to approximate the volatility and autocorrelation of R&D expenditure in the data. The simulated mean value
for the ratio of adoption expenditures over net output is about 1.53%, which accords with the data. Indeed, the expenditure
 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498 91

Table 3
Calibration of parameter values.

Parameters BL IACM CM IA RA

Preferences
0.95 0.95 0.95 0.95 0.95
5 5 5 5 5
5 5 5 5 15
Technology
0.20 0.25 0.25 0.20 0.20
0.65 0.70 0.70 0.65 0.65
0.09 0.09 0.09 0.09 0.09
a 0.10 0.10 0.10 0.10 0.10
8 8 8 8 8
0.60 0.60 0.60 0.60 0.60
0.166 1 0.166 1 0.166
1 0.8833 0.8833 0.8833
0.8 0.8 0.8
0.98 0.98 0.98 0.98 0.98
Taxes
s 0.035 0.035 0.035 0.035 0.035
c 0.40 0.40 0.40 0.40 0.40
ls 0.13 0.13 0.13 0.13 0.13
ded 0.1625 0.15 0.15 0.15 0.15
a 0.10 0.10 0.10 0.10 0.10
Exogenous shocks
A 0.95 0.95 0.95 0.95 0.95
x 0.40 0.40 0.40 0.40 0.40
0 0.145 0.145 0.145
1 0.968 0.968 0.968
1.8951 1.8951 1.8951
A 0.0085 0.0108 0.0109 0.0093 0.0122
x 0.20 0.20 0.20 0.20 0.20
0.15 0.15 0.15
Corr{ A , ln( )} 0.80

Notes: BL: baseline calibration. IACM: model with instant technology adoption and constant markup. CM: model with
constant markup. IA: model with instant technology adoption. RA: model with higher risk aversion.

share of total R&D over corporate output for the period 19602007 is 2.14%, and the expenditure share of development is
1.56%.
Krusell et al. (20 0 0) provide estimates for the elasticity of substitution between capital and skilled labor; our reported
value = 0.6 falls within their plausible range of estimates. The Cobb-Douglas specication would yield comparable levels
of volatility for the nancial variables. The volatilities of these nancial variables also remain quite insensitive to changes
in the elasticity parameter in the adjustment cost function (8). The TFP process (9) is adjusted to match the observed
volatilities of output and investment.
The markup process (10) is estimated from Compustat data from six subsamples corresponding to all (100%) companies
reporting R&D activities and the top 50%, 60%, 70%, 80%, 90% companies with the highest ratio of R&D expenditure over
total revenue. After taking logs and detrending, the estimated persistence parameter  in (10) lies within the range of
1
values [0.9434, 0.9827], and the estimated volatility parameter  lies within the range of values [0.1413, 0.3862]. These
estimates roughly remain the same when considering the group of high-tech companies with SIC codes: 281, 283, 284, 289,
357, 367, 381, and 384. In light of all this evidence, let us x 1 = 0.968, and = 0.15. These values5 are also consistent
with parameterizations of New Keynesian models with price rigidities. Smets and Wouters (2007, Table 4, p. 597) report an
autocorrelation parameter 1 = 0.90, but this parameter jumps to 1 = 0.97 when the degree of price stickiness is moved
to a minimal value.
The corporate income tax rate c is set to 0.40. The nal production tax rate s is set to 0.035 to match the ratio of
indirect taxes to value added in the corporate sector. The lump-sum tax rate ls is set to 0.13, and the fraction for deductions
ded = 0.1625. The depreciation allowance is set to a = 0.1. This tax structure approximates our BL model to the data in
several dimensions: (i) a steady-state value of dividends over value added equal to 8%, (ii) a ratio of total taxes to (after-tax)
dividends equal to 2.8, (iii) a volatility of taxes equal to 4.30%, and (iv ) a correlation coecient of taxes with corporate value
added equal to 0.69.

5
In a recent paper, Corhay et al. (2015) present an estimation of the aggregate markup as an inverse function of the labor income share. This indirect
procedure seems to yield more persistence than our computations of the markup measure.
92 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498

Fig. 2. Impulseresponse functions to a perturbation in A.

Notes: Response to a positive perturbation of A by one standard deviation from the deterministic steady state. The y-axis measures percentage deviation
from the deterministic steady-state value and the x-axis refers to annual dates. YN: corporate value added. C: consumption. I: investment. H (Z ):
adoption expenditures. MVC: market value of corporations. pk K: value of installed capital. PVT: present value of lump-sum taxes. V + : value of adopted
technologies. J + (Z ): option value of inventions currently available but not yet adopted. : present value of future inventions. D: dividends. PD: price-
dividend ratio. R: risk-free rate. RC: return of MVC. TAX: total taxes.

4.2. Impulse-response functions

Figs. 2 and 3 display impulse-response functions for the TFP index A, and the price markup for our baseline calibration
BL. Changes in the stock of available technologies Z will be discussed later.
An increment of one standard deviation over the deterministic steady-state value of A has a more persistent effect on
MVC than in the neoclassical growth model because of the extra investment in technology adoption H. An increase in TFP
stimulates consumption C and investment I, and dividends D go down. The interest rate R goes down to accommodate
convergence back to the steady state. The demand for intermediate goods is increased and prots per variety go up.
 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498 93

Fig. 3. Impulseresponse functions to a perturbation in

Notes: Response to a positive perturbation of by one standard deviation from the deterministic steady state. The y-axis measures percentage deviation
from the deterministic steady-state value and the x-axis refers to annual dates. YN: corporate value added. C: consumption. I: investment. H (Z ):
adoption expenditures. MVC: market value of corporations. pk K: value of installed capital. PVT: present value of lump-sum taxes. V + : value of adopted
technologies. J + (Z ): option value of inventions currently available but not yet adopted. : present value of future inventions. D: dividends. PD: price-
dividend ratio. R: risk-free rate. RC: return of MVC. TAX: total taxes.

Higher prots along with lower interest rates lead to increases in all components of MVC in Proposition 3.1 (i.e., V, J, ). An
increment of one standard deviation over the deterministic steady-state value of boosts MVC and D considerably, I and C
go down slightly, and H increases. The aggregate rm purchases less intermediate goods, and C and I go down because of
the decline in the productivities of capital and labor.
Note that the impact on MVC of is ve times greater than that of A. The change in A has similar effects on C, K and
MVC, whereas the change in leaves C and K almost unaffected. A positive change in A depresses D, and moves considerably
the price-dividend ratio PD in spite of the minor change in MVC. The change in produces a positive strong move in D, and
94 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498

Table 4
Standard deviations.

Data BL IACM CM IA RA

250 28 850 250 28 850 250 250 250 28 850 250 28 850

YN 3.57 2.06 2.91 3.44 1.28 3.19 3.42 3.47 3.48 1.29 3.23 3.45 1.72 2.99
(2.98, 4.16) (1.64, 2.48) (2.21, 3.61)
C 1.84 0.73 1.68 2.88 1.01 2.70 2.85 2.92 2.90 1.03 2.71 2.83 1.17 2.57
(1.45, 2.23) (0.61, 0.86) (1.26, 2.10)
I 9.26 4.65 7.97 12.17 7.16 9.84 7.87 7.99 11.02 5.67 9.45 11.81 8.49 8.21
(7.53, 10.99) (3.98, 5.32) (6.22, 9.72)
R&D 6.49 2.27 6.01 8.02 3.14 7.37 7.27 8.39 3.37 7.69
(4.99, 7.99) (1.89, 2.65) (4.68, 7.34)
MVC 24.93 11.96 21.83 16.30 7.38 14.53 2.38 2.78 14.32 6.35 12.84 15.66 6.53 14.23
(19.49, 30.37) (8.08, 15.83) (17.26, 26.40)
D 25.68 17.64 18.50 28.37 15.04 24.05 13.29 15.49 25.73 12.21 22.65 24.57 15.89 18.73
(18.76, 32.61) (12.23, 23.05) (14.72, 22.28)
PD 31.47 16.86 26.37 24.77 18.48 16.48 12.89 14.27 21.07 14.73 15.07 23.79 19.89 13.03
(23.67, 39.27) (13.36, 20.36) (18.62, 34.13)
RC 18.48 17.22 6.53 11.05 10.12 4.42 0.78 1.35 9.51 8.67 3.88 9.52 8.68 3.91
(11.94, 25.02) (11.48, 22.95) (4.59, 8.47)
R 2.38 1.43 1.87 1.44 1.22 0.77 0.21 0.21 1.07 0.88 0.60 1.41 1.20 0.73
(1.69, 3.06) (1.06, 1.80) (1.29, 2.44)
ER 18.69 17.53 6.27 10.81 9.71 4.73 0.77 1.35 9.33 8.38 4.10 9.29 8.28 4.21
(12.41, 24.98) (12.00, 23.05) (4.55, 7.99)
TAX 7.21 4.44 5.49 4.30 1.67 3.96 4.64 4.69 4.68 1.83 4.31 6.29 2.23 5.88
(5.52, 8.89) (3.18, 5.70) (4.35, 6.63)
MU75 5.35 1.74 5.03 3.50 1.45 3.19 3.50 1.45 3.19 3.61 1.43 3.32
(4.51, 6.18) (1.40, 2.07) (4.14, 5.91)

Notes: Standard deviations computed from log values for frequency bands of 250, 28, and 850 years. YN: corporate value added. C: consumption. I:
investment. R&D: R&D expenditure. MVC: market value of corporations. D: dividends. PD: price-dividend ratio. RC: return of MVC. R: risk-free rate. ER: excess
return of MVC over the risk-free rate. TAX: total taxes. MU75: average markup for the top 75% companies with the highest ratio of R&D expenditure over
total revenue. The corresponding variable in the model is . BL: baseline calibration. IACM: model with instant technology adoption and constant markup.
CM: model with constant markup. IA: model with instant technology adoption. RA: model with higher risk aversion. Newey-West corrected 95-percent
condence intervals are in parentheses. See Appendix for denitions and data sources.

dampens PD. This strong effect of on D comes from our simplistic modeling of the production of intermediate goods, and
will become evident in our further quantitative analysis.

4.3. Volatility

Simulated moments are obtained from equilibrium paths over 30 0 0 observations. Since all level variables are expressed
in log values, the standard deviation corresponds to the (relative) volatility. Both data and equilibrium paths have been
ltered with a band pass lter for a frequency band of 250 years, which is further decomposed into windows of 28 and
850 years.
Table 4 reports standard deviations for our baseline calibration BL and four other variants of this benchmark calibration.
For BL the volatility of real macroeconomic variables is quite similar to the prototypical real business-cycle model (Cooley
and Prescott, 1995). The standard deviations of nancial variables MVC and PD are over two-thirds of those found in the
data. Both in the data and in our model the volatility of RC (the return of MVC) is much higher than the volatility of the
risk-free rate R. Hence, the volatility of the excess return ER is driven by RC rather than by R. Our model generates an
adequate volatility for dividends D. In our BL calibration, the labor income share hovers around 63% and it is reasonably
volatile. Perturbations of the markup process affect the labor income share because of prot swings in the production of
intermediate goods. The model with variable labor does not signicantly improve upon the volatility of nancial variables.
Total taxes are less volatile than in the data; alternative calibrations show that taxes are not a relevant source of volatility.
Several variants of BL are included in Tables 3 and 4 to assess the importance of lags in technology adoption and ex-
ogenous markups. Production parameters and , and tax parameter ded are adjusted to keep an appropriate share of
dividends in value added.
IACM: A model with instant technology adoption and a constant markup. Lags in technology adoption are ruled out by
letting = 1, and exogenous uctuations in markups are ruled out by letting be equal to its steady-state value: = 1.18.
Note that in this case the volatility of nancial variables is roughly the same as that of real variables. Only the volatilities of
D and PD get sizable. As pointed out above, TFP shocks lead to a strong negative correlation of D with YN. The volatility of
PD arises from countercyclical shifts in D rather than procyclical shifts in MVC.
CM: A model with a constant markup. From our baseline calibration BL, let = 1.18. From Table 4, there are no notice-
able changes in the volatility of our economic variables for models IACM and CM. Hence, innovation and lags in technology
adoption do not contribute to increase the volatility of our nancial variables.
 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498 95

Table 5
Correlations.

Correlation with YN Correlation with D Correlation with MVC Autocorrelation

Data Model Data Model Data Model Data Model

YN 1 1 0.27 0.55 0.15 0.08 0.58 0.83

(0.48, 0.06) (0.18, 0.47) (0.41, 0.75)
C 0.70 0.99 0.23 0.47 0.04 0.10 0.80 0.85
(0.54, 0.84) (0.47, 0.01) (0.36, 0.26) (0.70, 0.89)
I 0.73 0.88 0.56 0.68 0.06 0.05 0.71 0.56
(0.58, 0.88) (0.76, 0.35) (0.43, 0.31) (0.57, 0.85)
R&D 0.41 0.01 0.06 0.19 0.51 0.40 0.83 0.83
(0.05, 0.76) (0.33, 0.45) (0.25, 0.78) (0.75, 0.92)
MVC 0.15 0.08 0.23 0.49 1 1 0.73 0.74
(0.18, 0.47) (0.15, 0.61) (0.57, 0.89)
D 0.27 0.55 1 1 0.23 0.49 0.55 0.64
(0.48, 0.06) (0.15, 0.61) (0.36, 0.77)
PD 0.34 0.58 0.63 0.82 0.60 0.09 0.69 0.27
(0.06, 0.62) (0.77, 0.49) (0.34, 0.86) (0.53, 0.86)
RC 0.28 0.13 0.44 0.42 0.32 0.37 0.07 0.15
(0.40, 0.16) (0.14, 0.75) (0.10, 0.54) (0.33, 0.17)
R 0.13 0.06 0.03 0.18 0.03 0.50 0.60 0.06
(0.32, 0.06) (0.17, 0.23) (0.35, 0.29) (0.41, 0.78)
ER 0.26 0.11 0.44 0.37 0.32 0.42 0.08 0.08
(0.39, 0.13) (0.12, 0.75) (0.10, 0.54) (0.32, 0.14)
TAX 0.61 0.69 0.14 0.11 0.20 0.39 0.53 0.81
(0.44, 0.77) (0.35, 0.06) (0.03, 0.43) (0.33, 0.73)
MU75 0.28 0.45 0.19 0.86 0.37 0.58 0.84 0.79
(0.01, 0.58) (0.07, 0.47) (0.11, 0.64) (0.76, 0.92)
V + 0.13 0.39 0.96 0.75
J + (Z ) 0.22 0.31 0.85 0.78
0.54 0.13 0.73 0.78

Notes: Selected correlations for model BL. These statistics are computed from log values for a frequency band of 250 years. YN: corporate value added. C:
consumption. I: investment. R&D: R&D expenditure. MVC: market value of corporations. D: dividends. PD: price-dividend ratio. RC: return of MVC. R: risk-
free rate. ER: excess return of MVC over the risk-free rate. TAX: total taxes. MU75: average markup for the top 75% companies with the highest ratio of R&D
expenditure over total revenue. The corresponding variable in the model is . V + : value of adopted technologies. J + (Z ): option value of inventions
currently available but not yet adopted. : present value of future inventions. Newey-West corrected 95-percent condence intervals are in parentheses.
See Appendix for denitions and data sources.

IA: A model with instant technology adoption. From baseline calibration BL, let = 1. In this case, standard deviations
are also reported over frequency bands of 28 and 850 years. Lags in technology adoption contribute to a rather small
increase in the volatility of nancial variables of about 10%. The most noticeable change is the volatility of the risk-free rate
R, which goes from 1.07% in the IA model to 1.44% in the BL model, whereas it is 2.48% in our data set. The volatility of R
in the data is expected to be higher because of unexpected ination.
RA: A model with higher risk aversion. The coecient of risk aversion now moves from  = 5 to  = 15, and the elas-
ticity of intertemporal substitution EIS 1/ = 1/5 remains unchanged. Also, we let Cor r A , ln( ) = 0.80, as opposed to
independent shocks in the BL calibration. Given that > , the representative household has a preference for early reso-
lution of uncertainty (cf. Kaltenbrunner and Lochstoer, 2010). As discussed in Section 4.2, the positive correlation between
shocks A and is essential to match the correlations of YN with markups, which appear with a negative bias in the BL
calibration; see Table 5. These changes do not substantially affect the volatility of our nancial variables including the risk-
free interest rate, but generate a more suitable risk premium of 1.59% as compared to 6.5% in our data (with a Newey-West
corrected standard error of 1.92%).
In summary, product price markups generate a signicant increase in the volatility of the nancial variables without
compromising the volatility of real aggregates. With instant technology adoption and constant markups the volatility of
MVC is equal to 2.38%, as opposed to 24.93 in the data. The introduction of a stochastic markup process rises the volatility
of MVC to 14.32%. Including lags in technology adoption increases the volatility of MVC to 16.30%, as well as the volatility of
the risk-free rate. Sizable equity premia require markups to be correlated with TFP.

4.4. Correlations

Another dimension of the excess volatility puzzle is the lack of correlation between real and nancial variables. These
correlations do not differ signicantly within the subgroup of models with stochastic markups, and hence the IA and RA
models will be omitted. Models IACM and CM lack economic interest because of their low volatility for the nancial vari-
ables. Hence, Table 5 displays some representative correlations for our BL model for the wider frequency band of 250 years.
96 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498

Correlations with YN: Real business-cycle models present high correlations of output with real and nancial variables.
Our baseline calibration BL inherits the high correlation of output with real variables, but recovers the low correlation of YN
with nancial variables; see Section 4.2. Markups are procyclical in the data, but negatively correlated with output in the
model. This is because stochastic innovations are uncorrelated in our BL calibration.
Correlations with D: Changes in TFP produce a rather strong negative correlation of dividends D with both real economic
activity and nancial variables. As we can see from Table 5 these correlations are much milder under our baseline calibration
BL. Still, there is a strong negative correlation between D and YN and between D and RC, which require a more suitable
correlation of the TFP and markup processes; i.e., see our discussion of the IACM and RA calibrations.
Correlations with MVC: While installed physical capital pkt kt+1 and the present value of lump-sum taxes PVTt can be
highly correlated with YN, C, and I, the remaining portion Vt+ t + Jt+ (Zt t ) + t of MVC in Proposition 3.1 exhibits a much
lower correlation.6 Hence, MVC is mildly correlated with several other economic aggregates. Our BL model generates a rather
low correlation of MVC with PD and a strong negative correlation of MVC with R. As seen in Table 5, in our model markups
are strongly correlated with D, which in turn affects the correlation of MVC with PD.
Autocorrelations: For our purposes, the autocorrelation coecient of dividends D is most critical. In both model and
data, this coecient is around 0.60. Our BL model captures well the autocorrelation of our economic variables, but fails for
PD and R. Actually, the problem is the autocorrelations of PD and R for the frequency band of 28 years.
It follows that our BL model mimics quite well several second-order moments. The main issues come from the negative
biases in the correlations of MVC with PD and MVC with R, and the autocorrelations of PD and R for the frequency band of
28 years. Apart from lessening the strong correlation between markups and dividends, this suggests the need for medium-
term time-varying risk (cf., Greenwald et al., 2014), which may come from monetary and nancial factors. In the recent
economic crisis, market sentiment and increased nancial risk along with a loss of collateral quality could be blamed for
downward shifts in MVC, PD, and R.

5. Variance decomposition of the price-dividend ratio PD

A vast literature contends that the variance of PD can be explained by changes in expectations of future returns (e.g.,
Campbell and Shiller, 1988; Cochrane, 1992). These results are not generally accepted. Ang (2002) and Ang and Bekaert
(2007) question their econometric signicance, whereas Boudoukh et al. (2007) and Larrain and Yogo (2008) propose broader
measures of protability accounting for a larger fraction of the variance of PD.
Following Campbell and Shiller (1988), the price-dividend
 ratio PD can be approximated under a log-linearization over
MVCt  MVCt +Dt
an observed sample path. Let pdt ln Dt , rct ln MVCt1 , and dt ln( DDt ). Then, for every horizon N the variance
t1
of pd can be decomposed into the following sources of volatility:
 N s1 rc
 
s=1
Cov Et t+s ,pdt
i Expected future returns: CVARN,rc Var ( pdt )
100
 N s1 d
 
Cov Et s=1 t+s ,pdt
ii Dividend growth: CVARN,d Var ( pdt )
100 and
Cov{Et [ N pdt+N ],pdt }
iii Terminal price component: CVARN,pd Var ( pdt )
100

where 0 < < 1. Table 6 provides a variance decomposition of pd for N = 30. As in Cochrane (1992), this table reports OLS
estimations of the forecastable variation of future returns over 30-year windows. Data statistics are calculated over our
annual set of observations for the time period 19482007, and model statistics are calculated over a simulated sample of
50,0 0 0 observations. Similar results are obtained when statistics are computed as averages over multiple simulated samples.
In the data, about 50% of the volatility of pd is explained by predictable variations in dividend growth. This relatively high
value seems quite reasonable for our broad measure of dividends (Larrain and Yogo, 2008). Note that our data estimates
present large standard errors, which is usually ascribed to the high autocorrelation observed in nancial variables (e.g., Ang,
20 02; Ang and Bekaert, 20 07; Cochrane, 20 08). In our BL model, about three-fourths of the variance of pd is accounted by
dividends; these values are close but lie outside the 95-percent condence intervals reported in Table 6. Of course, some
positive bias is to be expected since our model is just made up of a representative household, and is abstracting from some
other sources of volatility. The other two models with stochastic markups (IA and RA) present similar variance decomposi-
tions. The models with constant markup (IACM and CM) generate higher values for CVAR30,rc , but they lack economic interest
because the volatility of MVC is implausibly low.
Finally, Table 7 reports the predictive power of pd over future excess returns and dividend growth. Because of our broad
measure of dividends, it is worth noting that in our data the price-dividend ratio pd has some explanatory power not only
for future excess returns but also for dividend growth. Our BL and RA calibrations present similar patterns for dividends, and
lack explanatory power for future excess returns; more precisely, the coecients for future excess returns display the right

6
In our simulations the value of installed physical capital pkt kt+1 ranges between 22 and 63% of the tax-adjusted MVC (i.e., qt + PV Tt ), the value of
existing technologies Vt+ t ranges between 30 and 65%, whereas the option value of adopting future technologies Jt+ (Zt t ) + t is much smaller and
lies between 10 and 25%. These gures seem quite plausible. Hall (2001) argues that in periods of high technological activity the weight of capital may get
down to one fourth of its peak value.
 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498 97

Table 6
Variance decomposition of pd.

CVAR30,rc CVAR30,d CVAR30,pd Total

Data 38.86 49.84 8.80 97.51

(24.65, 53.06) (24.91, 74.78) (-5.53, 23.13)
BL 20.16 73.99 6.15 100.32
IACM 41.43 42.95 10.54 94.92
CM 37.98 47.79 10.44 96.21
IA 18.98 72.18 7.20 98.36
RA 19.58 76.98 3.20 99.76
 
Cov{Et [ 30
s=1
s1
rct+s ],pdt } Cov{Et [ 30
s=1
s1
dt+s ],pdt }
Notes: Data and model OLS estimates of CVAR30,rc Var ( pdt )
100, CVAR30,d Var ( pdt )

Cov{Et [ 30 pdt+30 ],pdt } MVCt +Dt 
100, and CVAR30,pd . Total refers to the sum of these three statistics. Here, rc t ln ,  d t

Var ( pdt ) MVCt1
ln( DDt1
t
), and pdt ln MVC
Dt
t
. MVC: market value of corporations. D: dividends. BL: baseline calibration. IACM: model
with instant technology adoption and constant markup. CM: model with constant markup. IA: model with instant tech-
nology adoption. RA: model with higher risk aversion. Newey-West corrected 95-percent condence intervals are in
parentheses. See Appendix for denitions and data sources.

Table 7
Excess return and dividend predictability.

Data BL IACM CM IA RA

er d er d er d er d er d er d

Horizon in Years: N=1
0.222 0.101 0.007 0.288 0.0 0 0 0.053 0.001 0.057 0.001 0.228 0.017 0.520
(0.340, 0.104) (0.082, 0.286)
2
R 0.168 0.021 0.0 0 0 0.225 0.0 0 0 0.032 0.001 0.032 0.0 0 0 0.180 0.003 0.396
Horizon in Years: N=3
0.124 0.137 0.003 0.126 0.0 0 0 0.049 0.002 0.055 0.0 0 0 0.110 0.016 0.196
(0.183, 0.065) (0.029, 0.245)
R2 0.251 0.155 0.0 0 0 0.203 0.001 0.091 0.006 0.093 0.0 0 0 0.170 0.008 0.352
Horizon in Years: N=5
0.106 0.079 0.0 0 0 0.093 0.0 0 0 0.044 0.001 0.049 0.003 0.086 0.012 0.125
(0.152, 0.060) (0.021, 0.137)
R2 0.357 0.155 0.0 0 0 0.220 0.001 0.135 0.009 0.140 0.0 0 0 0.192 0.008 0.321
Horizon in Years: N=7
0.085 0.052 0.001 0.076 0.0 0 0 0.040 0.002 0.045 0.004 0.073 0.011 0.092
(0.122, 0.047) (0.016, 0.089)
2
R 0.352 0.109 0.0 0 0 0.228 0.001 0.170 0.010 0.178 0.002 0.209 0.009 0.289
Horizon in Years: N=10
0.087 0.033 0.001 0.062 0.0 0 0 0.036 0.001 0.040 0.004 0.061 0.009 0.073
(0.122, 0.051) (0.018, 0.084)
R2 0.395 0.056 0.0 0 0 0.244 0.002 0.217 0.013 0.226 0.003 0.230 0.009 0.289

Notes: Excess returns and dividend growth are regressed on the price-dividend ratio. The table reports predictive coecients and R statistics for 2
 N s = N
regressions: (1/N ) ss==1 ert+s = + pdt and (1/N ) s=1 dt+s = + pdt . Here, ert ln (RCt ) ln (Rt ), dt ln ( Dt1 ), and pdt ln
Dt MVCt
Dt
. RC: return of
MVC. R: risk-free rate. D: dividends. MVC: market value of corporations. BL: baseline calibration. IACM: model with instant technology adoption and constant
markup. CM: model with constant markup. IA: model with instant technology adoption. RA: model with higher risk aversion. Newey-West corrected 95-
percent condence intervals are in parentheses. See Appendix for denitions and data sources.

sign but are not statistically signicant. Again, this lack of predictability conrms that our BL calibration may be missing
some sources of medium-term time-varying risk.

6. Concluding remarks

From the Standard & Poors Compustat database, popular accounting measures of protability are weakly correlated with
stock values, whereas product price markup measures exhibit more volatility and are highly correlated with stock values.
The predictive power of markups is enhanced when companies are grouped by R&D intensity.
Our two-sector model has been instrumental to explore the inuence of technology innovation, TFP shocks, and product
price markups on asset price volatility. Overall, the model generates a volatility for the market value of corporations of
about 16.30% as opposed to 24.93% in the data without imposing further volatility on the real economy. As compared
to TFP shocks, exogenous variations in product price markups lead to about a ve-fold increase in long-term asset price
volatility with relatively little impact on the real economy. Innovation and lags in technology adoption have rather mild
effects on the volatility of our nancial variables.
In our model, uctuations in price markups are strongly correlated with dividends because of a simplistic modeling of the
production of intermediate goods. This results in a low correlation between the stock market value and the price-dividend
ratio, as well as a low autocorrelation of the price dividend-ratio. Also, our baseline calibration is unable to account for the
98 M.A. Iraola, M.S. Santos / Journal of Monetary Economics 90 (2017) 8498

equity premium. A sizable equity premium and suitable stock market predictability may be obtained with correlated shocks
and high degrees of risk aversion. Finally, the models performance may be improved by considering time-varying risk and
monetary factors to generate higher correlation among nancial variables and persistent interest rates.

Supplementary material

Supplementary material associated with this article can be found, in the online version, at 10.1016/j.jmoneco.2017.07.002.

References

Alburquerque, R.A., Eichenbaum, M., Luo, V.X., Rebelo, S.T., 2016. Valuation risk and asset pricing. J. Finance 6, 28612904.
Ang, A., 2002. Characterizing the ability of dividend yields to predict future dividends in log-linear present value models. Unpublished paper.
Ang, A., Bekaert, G., 2007. Stock return predictability: is it there? Rev. Financ. Stud. 20, 651707.
Barlevy, G., 2007. On the cyclicality of research and development. Am. Econ. Rev. 97, 11311164.
Barsky, R.B., DeLong, J.B., 1993. Why does the stock market uctuate? Q. J. Econ. 108, 291312.
Bena, J., Garlappi, L., 2013. Corporate innovation and returns. Unpublished paper.
Boldrin, M., Peralta-Alva, A., 2009. What happened to the U.S. stock market? Accounting for the past 50 years. Federal Reserve Bank St. Louis Rev. 91,
627646.
Boudoukh, J., Michaely, R., Richardson, M., Roberts, M.R., 2007. On the importance of measuring payout yield: implications for empirical asset pricing. J.
Finance 62, 877915.
Campbell, J.Y., Ammer, J., 1993. What moves the stock and bond markets? A variance decomposition for long-term asset returns. J. Finance 48, 337.
Campbell, J.Y., Shiller, R.J., 1988. Stock prices, earnings, and expected dividends. J. Finance 43, 661676.
Cochrane, J.H., 1992. Explaining the variance of price-dividend ratios. Rev. Financ. Stud. 2, 243280.
Cochrane, J.H., 2008. The dog that did not bark: a defense of return predictability. Rev. Financ. Stud. 21, 15331575.
Comin, D., Gertler, M., 2006. Medium-term business cycles. Am. Econ. Rev. 96, 523551.
Cooley, T.F., Prescott, E.C., 1995. Economic growth and business cycles. In: Cooley, T.F. (Ed.), Frontiers of Business Cycle Research. N.J.: Princeton University
Press, pp. 138.
Corhay, A., Kung, H., Schmid, L., 2015. Competition, markups, and predictable returns. Unpublished paper.
Croce, M.M., 2014. Long-run productivity: a new hope for production-based asset pricing? J. Monet. Econ. 66, 1331.
Fama, E.F., French, R.F., 2006. Protability, investment and average returns. J. Financ. Econ. 82, 491518.
Gabaix, X., 2011. The granular origins of aggregate uctuations. Econometrica 79, 732772.
Grleanu, N.B., Panageas, S., Yu, J., 2012. Technological growth and asset pricing. J. Finance 67, 12651292.
Gomes, J., Kogan, L., Zhang, L., 2003. Equilibrium cross section of returns. J. Political Econ. 111, 693732.
Gomme, P., Ravikumar, B., Rupert, P., 2011. The return to capital and the business cycle. Rev. Econ. Dyn. 14, 262278.
Gourio, F., 2012. Disaster risk and business cycles. Am. Econ. Rev. 102, 27342766.
Greenwald, D.L., Lettau, M., Ludvigson, S.C., 2014. Origins of stock market uctuations. NBER. Working Paper no. 19818.
Grullon, G., Hund, J., Weston, J., 2013. A granular analysis of corporate investment. Unpublished paper.
Hall, B.H., 2007. Measuring the returns to R&D: The depreciation problem. NBER Working Paper no. 13473.
Hall, R.E., 2001. The stock market and capital accumulation. Am. Econ. Rev. 91, 11851202.
Jaimovich, N., Floetotto, M., 2008. Firm dynamics, markup variations and the business cycle. J. Monet. Econ. 55, 12381252.
Jermann, U.J., 1998. Asset pricing in production economies. J. Monet. Econ. 41, 257275.
Kaltenbrunner, G., Lochstoer, L.A., 2010. Long-run risk through consumption smoothing. Rev. Financ. Stud. 23, 31903224.
Kogan, L., Papanikolaou, D., 2014. Growth opportunities, technology shocks, and asset prices. J. Finance 69, 675718.
Krusell, P., Ohanian, L.E., Rios-Rull, J.V., Violante, G.L., 20 0 0. Capital-skill complementarity and inequality: a macroeconomic analysis. Econometrica 68,
10291054.
Kung, H., Schmid, L., 2015. Innovation, growth, and asset prices. J. Finance 70, 10011037.
Larrain, B., Yogo, M., 2008. Does rm value move too much to be justied by subsequent changes in cash ow. J. Financ. Econ. 87, 200226.
McGrattan, E.R., Prescott, E.C., 2005. Taxes, regulations, and the value of U.S. and U.K. corporations. Rev. Econ. Stud. 72, 767796.
Nekarda, C.J., Ramey, V.A., 2013. The cyclical behavior of the price-cost markup. NBER Working Paper no. 19099.
Novy-Marx, R., 2013. The other side of value: the gross protability premium. J. Financ. Econ. 108, 128.
Peralta-Alva, A., 2007. The information technology revolution and the puzzling trends in Tobins average q. Int. Eco. Rev. 48, 929951.
Romer, P.M., 1990. Endogenous technological change. J. Political Econ. 98, 71102.
Santos, M.S., 1999. Numerical solution of dynamic stochastic models. In: Taylor, J., Woodford, M. (Eds.), Handbook of Macroeconomics, Elsevier, Amsterdam,
pp. 311386.
Schmitt-Groh, S., Uribe, M., 2004. Solving dynamic general equilibrium models using a second-order approximation to the policy function. J. Econ. Dyn.
Control 28, 755775.
Smets, F., Wouters, R., 2007. Shocks and frictions in US business cycles: a bayesian DSGE approach. Am. Econ. Rev. 97, 586606.
Wright, S., 2004. Measures of stock market value and returns for the U.S. nonnancial corporate sector, 190 020 02. Rev. Income Wealth 50, 561584.